An Adaptive System for Detecting Driving Abnormality of Individual Drivers Using Gaussian Mixture Model

Abnormal driving of a driver often leads to some undesirable situations. To avoid such circumstances, this paper proposes a driver-adaptive system that can detect abnormality within one’s own driving style. To achieve such a feat, firstly, a soft clustering approach is proposed to determine the driving style in three categories - aggressive, moderate, and protective. Then, it proposes an algorithm to analyze the driving profile and detect the variation of abnormality of the driver. Statistics of different vehicle kinetic characteristics, i.e., velocity, space and time headway, acceleration, jerk factor, etc., are considered for the clustering process. Maximal Information Coefficient (MIC) is used to select the important features. Analyzing the results of different clustering algorithms such as Gaussian Mixture Model (GMM), FCM, and K-means, we used GMM in the proposed system for determining real-time driving style. The actual driving style is determined using the long-term driving history of the subject vehicle. Using GMM’s soft clustering properties, the current deviation is measured by comparing the real-time driving style to the actual style. A learning factor is accommodated to update the actual style in every cycle so that the proposed system can adapt to the driver’s skill-changing state. Real-life driving depends on real-time human decisions. Therefore, we have used the reconstructed NGSIM I-80 real-life traffic data to build and validate the proposed system.